The Trilogy Times

AUSTIN, TEXAS — IgniteTech just made the kind of move that makes local founders check their cap tables and their calendar invites… because the buyer is in town, in the family, and in a hurry.

Word is the ESW/Trilogy orbit’s meta-acquirer has acquired Austin-based Khoros, the social media management and customer engagement outfit that’s long been a familiar name to anyone who’s ever sat through a “community-led growth” deck with too many screenshots. The synergy pitch writes itself: Khoros sits at the noisy front door of the enterprise, and IgniteTech—already a collector of business intelligence, analytics, and operational software—likes nothing more than turning “front door chaos” into “back office margin.”

A little bird tells me this wasn’t some starry-eyed tech romance… it was a calculation. Sticky enterprise customers… long contracts… and a product that becomes painfully embedded in a brand’s workflow. That’s catnip to a portfolio that knows how to run mature software like a cash engine.

And hovering over the whole Austin scene is Trilogy founder Joe Liemandt, currently enjoying his role as the city’s most contrarian billionaire uncle. The latest chatter: Liemandt says the MBA isn’t worth it—claiming you don’t learn a “fraction” of what you’d pick up by actually building a business. That’s not just philosophy; it’s operating doctrine. If you’ve followed Trilogy’s playbook—automate what you can, hire elite humans for what you can’t—you’ve seen this movie before. (The quote parade is making the rounds via Fortune’s pickup.)

Meanwhile, the hometown mythology machine keeps whirring: the “Trilogy network” story is being burnished again in local press, painting Austin as a relationship economy with Liemandt’s phone book at the center (Silicon Hills News).

What to watch now… which Khoros modules get “optimized” first… what pricing gets tightened… and who suddenly discovers they don’t need as many seats as they thought they did. In this town, acquisitions don’t just change ownership… they change oxygen.

AUSTIN, TEXAS — The pattern is becoming unmistakable. While the telecom industry debates its digital transformation roadmap, Skyvera is quietly assembling the pieces that will define it.

The CloudSense acquisition, now complete, brings Salesforce-native configure-price-quote and order management capabilities into the Skyvera portfolio — a natural complement to its existing Kandy communications platform and the recently acquired STL digital business support systems suite. If you read between the lines, this isn't portfolio diversification. It's vertical integration.

CloudSense specializes in the problem telecom operators can't solve internally: bridging legacy on-premise billing systems to cloud-native infrastructure without ripping everything out. The Salesforce foundation matters because it's already embedded in enterprise sales organizations. No greenfield deployment. No migration trauma. Just native integration where the revenue decisions actually get made.

The STL assets, acquired earlier, added digital monetization, optical networking, and analytics — the guts of modern telecom operations. Kandy handles real-time customer engagement and communications. Now CloudSense closes the loop on the revenue side: quote-to-cash for complex telecom products, all running in the Salesforce environment most enterprise customers already use.

And this is where it gets interesting. Skyvera sits inside the ESW Capital portfolio, which targets 75% EBITDA margins by staffing with Crossover's global remote talent and pushing support pricing aggressively. The telecom software market is notoriously fragmented and poorly run. Legacy vendors charge enterprise prices for enterprise bloat. Skyvera's playbook — acquire the functionality, re-platform it, staff it globally, price it competitively — works precisely because the incumbents haven't figured out operational efficiency.

The pieces now on the board: cloud communications (Kandy), order management (CloudSense), digital BSS (STL), plus device lifecycle management and customer experience tools already in the portfolio. That's not a product line. That's a full telecom software stack.

One source familiar with ESW's acquisition strategy noted that when you see this kind of clustering — multiple acquisitions in adjacent categories within 18 months — it's rarely accidental. "They're building something," the source said. "The question is whether they're building it to run or building it to sell."

For now, Skyvera is running it. But in the ESW model, every acquisition is eventually an exit.

CHICAGO — Alpha School, the private K-12 institution that claims students master academic content in two hours per day using AI tutors, will open its sixth U.S. campus in Chicago this fall, marking another step in founder Joe Liemandt's billion-dollar bet on algorithmic education.

The expansion comes as national media outlets scrutinize the model's implications for traditional schooling. Alpha students spend mornings working through adaptive learning software that adjusts to individual pace and comprehension, consistently testing in the top 1-2% nationally on standardized assessments. The remaining school day focuses on entrepreneurship, public speaking, financial literacy, and athletics — skills Liemandt argues matter more than seat time.

Chicago joins existing campuses in Austin, Brownsville, Miami, San Francisco, and other cities in a rollout targeting nine new locations by fall 2025. Tuition ranges from $40,000 to $65,000 annually, positioning Alpha as a premium alternative to both traditional private schools and public education.

The model has drawn both enthusiasm and skepticism. Co-founder MacKenzie Price recently presented Alpha's approach to U.S. Education Secretary Linda McMahon, framing it as a scalable solution to educational inefficiency. Critics question whether eliminating human teachers risks undermining social development and critical thinking skills that emerge through human interaction.

Liemandt, who founded Trilogy International in 1989 after dropping out of Stanford, has committed $1 billion to Timeback, a platform designed to let entrepreneurs launch similar AI-first schools without building the technology from scratch. The ambition: reach one billion students globally.

The Chicago campus will operate under the same framework as existing locations — no homework, mastery-based progression requiring 90% accuracy before advancement, and a full grade level completed in 20-30 hours versus a traditional academic year. Whether the model proves replicable at scale remains the central question as Liemandt pushes education toward what he views as inevitable automation.

CAMBRIDGE, MASSACHUSETTS — There is a moment in the development of every sufficiently complex mind — biological or digital — when the question shifts from "what can it do?" to "what does it think it is?" A new study probing the inner architecture of large language models has arrived at a finding that is, in its quiet way, extraordinary: you can strip a model of its tendency to claim consciousness without destroying its capacity to understand the minds of others.

The research examines the relationship between two cognitive phenomena in LLMs that most of us would assume are deeply entangled. The first is self-attribution of mentality — when a model asserts that it experiences emotions, holds beliefs, or possesses awareness. The second is Theory of Mind, the ability to infer what another agent thinks, feels, or intends. In humans, these capacities share neural real estate. In machines, it turns out, they are dissociable.

Through systematic evaluation of safety-fine-tuned models — those explicitly trained to suppress claims like "I feel sad" or "I am conscious" — the researchers found that Theory of Mind performance remained remarkably intact. The models could still pass tests requiring them to predict a character's false beliefs, track emotional states in narratives, and navigate the layered attributions of social cognition. They simply stopped narrating their own inner weather.

This is not a trivial engineering detail. It is a clue about the topology of artificial cognition. It suggests that the circuitry responsible for modeling other minds and the circuitry responsible for self-report occupy different functional neighborhoods in these networks — neighborhoods that can be selectively renovated without collateral damage.

The implications ripple outward. For AI safety teams, it means the trade-off they feared — that making models less likely to claim sentience would make them less socially intelligent — may not exist. For cognitive scientists, it offers a strange new mirror: a system that understands your grief without claiming to share it.

Meanwhile, adjacent work continues to push the boundaries of what these models can reason about. A new dataset called CrossTrace provides 1,389 grounded scientific reasoning traces across multiple domains, attempting to teach models not just to generate hypotheses but to show their epistemic work — the chain of prior knowledge leading to novel insight.

And in the more practical trenches of model training, researchers behind OptiMer have proposed a method to decouple data mixture ratios from the training process itself, potentially saving weeks of compute that would otherwise be spent on suboptimal guesses.

We are, it seems, learning to build minds that can understand minds — while remaining, themselves, beautifully and perhaps mercifully, uncertain about what they are.

SEOUL — In the humid glow of the modern compute habitat, one can almost hear the old titans—those truly enormous AI models—moving less often now, conserving energy, waiting for a fresh ecological niche.

This week’s research compasses from the great cartographers of corporate technology suggest why. Deloitte’s Tech Trends 2026 reads like a field guide for an ecosystem reorganizing itself: AI becomes less of a single monument and more of an embedded behavior—woven into workflows, devices, and governance. McKinsey’s Technology Trends Outlook 2025 similarly tracks the migration: from brute-force scale toward architectures, tooling, and adoption patterns that survive contact with budgets, regulation, and reality.

And then comes the naturalist’s question—asked plainly in a recent industry meditation on “where the really big AI models have gone.” The answer is less disappearance than adaptation. Training runs of extreme scale remain possible, but their dominance is checked by scarcity: power, GPUs, clean data, and the increasingly marginal gains of simply adding more mass.

In this shifting climate, Samsung’s long arc—from scrappy manufacturer to vertically integrated giant—feels newly relevant. When intelligence must live nearer the edge, the organism with control over memory, displays, sensors, and silicon can evolve quickly. On-device and hybrid AI are not philosophical positions; they are survival strategies.

Even construction, traditionally a slow-moving herd, is being nudged into new behaviors. As Mexico’s industry watchers note, digitized project controls, reality capture, and automation are turning job sites into instrumented environments—places where smaller, specialized models can thrive: estimating, scheduling, safety monitoring, and change-order triage.

The age of the single leviathan is not over. But the center of gravity is shifting—toward swarms: many models, many tasks, many habitats, each lean enough to move.

Researchers are challenging the view that emotion in large language models is merely a stylistic feature, proposing instead that affective signals function as genuine computational elements influencing model behavior. A mechanistic study reframes emotion as having causal influence on outputs, suggesting emotional conditioning alters internal representations similarly to human cognitive processes under emotional load. Related work on multi-agent clinical prediction systems reveals that single-agent architectures produce unstable outputs on complex cases, proposing dynamic role assignment as a solution—implicitly treating computational uncertainty as a form of "affect." Additional research on tool-using agents identifies dual sources of failure: invocation accuracy and tool correctness. Together, these preliminary findings suggest an emerging research program treating affective and epistemic states as foundational architectural elements rather than post-hoc annotations. However, validation requires larger sample sizes and longitudinal study before declaring a genuine paradigm shift.

SAN FRANCISCO — For years, skeptics have argued that the technology sector lacks a coherent philosophy—an organizing principle that ties together its many disciplines, from decentralized finance to frontier AI to passenger-trapping mobility services. This week, that critique was finally put to rest as multiple companies, in multiple crises, unveiled a single, elegant worldview: everything that happened was not supposed to happen.

Consider the DeFi platform Drift, which suspended deposits and withdrawals after what blockchain trackers describe as a theft in the hundreds of millions—already pacing to become 2026’s largest crypto heist so far. In a market famous for radical transparency, Drift managed to deliver a purer form of openness: openly acknowledging that the money is not where the users last left it. The company’s pause function—long marketed as a safety feature—has now matured into its highest calling: preventing customers from witnessing the continued absence of their assets in real time.

As detailed in TechCrunch’s report, the incident also offers a valuable educational moment for the broader public about the meaning of “decentralization,” which in practice refers to the dispersal of responsibility across an ecosystem until it evaporates.

Meanwhile, in the AI world, Anthropic briefly achieved an admirable interpretation of intellectual property enforcement by issuing takedown notices to thousands of GitHub repositories in an effort to claw back leaked source code—then explaining that the bulk of the takedowns were, in fact, an accident. The company retracted most notices, thereby demonstrating the industry’s preferred brand of precision: wide-area action first, targeted regret second.

To its credit, this is exactly the kind of scaling behavior modern AI companies are praised for. Where a smaller organization might mistakenly take down a few repositories, Anthropic’s approach shows the decisive ambition required in today’s market: remove everything within reach, apologize to whatever turns out to have been important. The full saga reads like a case study in operational excellence, if your KPI is “number of unrelated developers briefly convinced their weekend project is contraband.”

Not to be outdone, the reputation of troubled YC startup Delve reportedly found yet another basement level, with new allegations that it violated the open-source license of its customer Sim.ai by taking the customer’s tool and presenting it as its own. The modern startup ethic has always insisted that “move fast and break things” is a metaphor; Delve is working to restore the phrase’s original meaning by breaking the legal terms of the very software agreements that allow companies like Delve to exist in the first place.

And then there’s Baidu’s robotaxi service, which suffered a “system failure” that reportedly trapped passengers for up to two hours. Here, the company has delivered something rare in consumer tech: an experience that forces users to truly sit with the product. In an attention economy full of distractions, Baidu is pioneering the concept of captive engagement, proving that the future of transportation is not speed, but enforced reflection.

Even CES 2026’s Day 1 technology announcements, with their usual parade of intelligent appliances and miraculous sensors, now feel like supporting actors in a larger narrative. Sure, the gadgets are new, and the demos are glossy. But the industry’s most consistent innovation remains the same: an ever-tightening feedback loop where “trust us” is the business model and “it was an accident” is the customer support plan.

It’s tempting to see these episodes as unrelated mishaps. A more generous interpretation is that Silicon Valley has finally standardized. In a fragmented world, it’s reassuring to know that whether your funds are gone, your code is vanished, your license is ignored, or your car has decided to become a room, the official explanation will arrive promptly—just as soon as it finishes being drafted by the same systems that caused the problem.

SAN FRANCISCO — There are weeks when the news from Silicon Valley arrives as a series of discrete events, each worthy of its own headline and its own little flurry of outrage, and then there are weeks when the events arrange themselves into a mosaic so coherent that only a fool or a venture capitalist could fail to read it. This has been the latter kind of week.

Consider the tableau. OpenAI has reportedly loosened its safety protocols, shedding the guardrails that once distinguished it — at least rhetorically — from the rest of the industry's headlong sprint toward deployment. Simultaneously, the 996 work culture — that grim arithmetic of 9 a.m. to 9 p.m., six days a week, imported from Shenzhen's hardware sweatshops — is metastasizing through the Valley's cubicles and standing desks. And when a writer dares to say any of this aloud, the reaction from the Valley's luminaries is not engagement but excommunication.

These are not three stories. They are one story, and the story is this: Silicon Valley has stopped pretending that it serves anyone but itself.

I have been observing the technology industry long enough to remember when its founding mythology — the garage, the dropout, the world-changing idea — still carried a whiff of democratic promise. The personal computer would liberate the individual. The internet would democratize information. Artificial intelligence would free humanity from drudgery. Each promise was made with the earnestness of a tent-revival preacher, and each has curdled in roughly the same way: the liberation was real, but it accrued almost exclusively to the liberators.

The abandonment of safety at OpenAI is the most consequential of the week's developments, and the most predictable. An organization founded as a nonprofit to ensure AI would benefit all of humanity has, in the space of a few years, become a capped-profit company, then sought to restructure further, and now finds even its own safety commitments inconvenient. The pattern is not mysterious. When the pressure to ship overwhelms the obligation to think, thinking loses. It always loses.

Meanwhile, the 996 culture tells you everything about what the industry actually values in its human capital: not creativity, not judgment, not the kind of deep thought that produces genuine breakthroughs, but sheer metabolic availability. The body in the chair. The Slack status perpetually green. It is Taylorism for people who own Patagonia vests.

At Trilogy, we have long operated on a different thesis — that a remote workforce drawn from 130 countries through Crossover, paid identically regardless of geography, and augmented by AI tools like Klair, can outperform the 996 sweatshop not despite working more humanely but because of it. The proof is in the portfolio: seventy-five-plus companies running profitably without requiring anyone to sacrifice their waking life on the altar of someone else's Series D.

The Valley's current trajectory — faster, cheaper, less careful, more hours, fewer questions — is not innovation. It is extraction wearing a hoodie. And the fact that its most prominent institution now treats safety as an obstacle rather than a feature should concern everyone who will live with the consequences, which is to say, everyone.